AI RESEARCH
ALARA for Agents: Least-Privilege Context Engineering Through Portable Composable Multi-Agent Teams
arXiv CS.AI
•
ArXi:2603.20380v1 Announce Type: cross Industry practitioners and academic researchers regularly use multi-agent systems to accelerate their work, yet the frameworks through which these systems operate do not provide a simple, unified mechanism for scalably managing the critical aspects of the agent harness, impacting both the quality of individual human-agent interactions and the capacity for practitioners to coordinate toward common goals through shared agent infrastructure.